1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (C) 2015 - ARM Ltd
4 * Author: Marc Zyngier <marc.zyngier@arm.com>
7 #include <linux/arm-smccc.h>
8 #include <linux/kvm_host.h>
9 #include <linux/types.h>
10 #include <linux/jump_label.h>
11 #include <uapi/linux/psci.h>
13 #include <kvm/arm_psci.h>
15 #include <asm/barrier.h>
16 #include <asm/cpufeature.h>
17 #include <asm/kprobes.h>
18 #include <asm/kvm_asm.h>
19 #include <asm/kvm_emulate.h>
20 #include <asm/kvm_hyp.h>
21 #include <asm/kvm_mmu.h>
22 #include <asm/fpsimd.h>
23 #include <asm/debug-monitors.h>
24 #include <asm/processor.h>
25 #include <asm/thread_info.h>
27 /* Check whether the FP regs were dirtied while in the host-side run loop: */
28 static bool __hyp_text update_fp_enabled(struct kvm_vcpu *vcpu)
31 * When the system doesn't support FP/SIMD, we cannot rely on
32 * the _TIF_FOREIGN_FPSTATE flag. However, we always inject an
33 * abort on the very first access to FP and thus we should never
34 * see KVM_ARM64_FP_ENABLED. For added safety, make sure we always
37 if (!system_supports_fpsimd() ||
38 vcpu->arch.host_thread_info->flags & _TIF_FOREIGN_FPSTATE)
39 vcpu->arch.flags &= ~(KVM_ARM64_FP_ENABLED |
42 return !!(vcpu->arch.flags & KVM_ARM64_FP_ENABLED);
45 /* Save the 32-bit only FPSIMD system register state */
46 static void __hyp_text __fpsimd_save_fpexc32(struct kvm_vcpu *vcpu)
48 if (!vcpu_el1_is_32bit(vcpu))
51 vcpu->arch.ctxt.sys_regs[FPEXC32_EL2] = read_sysreg(fpexc32_el2);
54 static void __hyp_text __activate_traps_fpsimd32(struct kvm_vcpu *vcpu)
57 * We are about to set CPTR_EL2.TFP to trap all floating point
58 * register accesses to EL2, however, the ARM ARM clearly states that
59 * traps are only taken to EL2 if the operation would not otherwise
60 * trap to EL1. Therefore, always make sure that for 32-bit guests,
61 * we set FPEXC.EN to prevent traps to EL1, when setting the TFP bit.
62 * If FP/ASIMD is not implemented, FPEXC is UNDEFINED and any access to
63 * it will cause an exception.
65 if (vcpu_el1_is_32bit(vcpu) && system_supports_fpsimd()) {
66 write_sysreg(1 << 30, fpexc32_el2);
71 static void __hyp_text __activate_traps_common(struct kvm_vcpu *vcpu)
73 /* Trap on AArch32 cp15 c15 (impdef sysregs) accesses (EL1 or EL0) */
74 write_sysreg(1 << 15, hstr_el2);
77 * Make sure we trap PMU access from EL0 to EL2. Also sanitize
78 * PMSELR_EL0 to make sure it never contains the cycle
79 * counter, which could make a PMXEVCNTR_EL0 access UNDEF at
80 * EL1 instead of being trapped to EL2.
82 write_sysreg(0, pmselr_el0);
83 write_sysreg(ARMV8_PMU_USERENR_MASK, pmuserenr_el0);
84 write_sysreg(vcpu->arch.mdcr_el2, mdcr_el2);
87 static void __hyp_text __deactivate_traps_common(void)
89 write_sysreg(0, hstr_el2);
90 write_sysreg(0, pmuserenr_el0);
93 static void activate_traps_vhe(struct kvm_vcpu *vcpu)
97 val = read_sysreg(cpacr_el1);
99 val &= ~CPACR_EL1_ZEN;
102 * With VHE (HCR.E2H == 1), accesses to CPACR_EL1 are routed to
103 * CPTR_EL2. In general, CPACR_EL1 has the same layout as CPTR_EL2,
104 * except for some missing controls, such as TAM.
105 * In this case, CPTR_EL2.TAM has the same position with or without
106 * VHE (HCR.E2H == 1) which allows us to use here the CPTR_EL2.TAM
107 * shift value for trapping the AMU accesses.
112 if (update_fp_enabled(vcpu)) {
113 if (vcpu_has_sve(vcpu))
114 val |= CPACR_EL1_ZEN;
116 val &= ~CPACR_EL1_FPEN;
117 __activate_traps_fpsimd32(vcpu);
120 write_sysreg(val, cpacr_el1);
122 write_sysreg(kvm_get_hyp_vector(), vbar_el1);
124 NOKPROBE_SYMBOL(activate_traps_vhe);
126 static void __hyp_text __activate_traps_nvhe(struct kvm_vcpu *vcpu)
130 __activate_traps_common(vcpu);
132 val = CPTR_EL2_DEFAULT;
133 val |= CPTR_EL2_TTA | CPTR_EL2_TZ | CPTR_EL2_TAM;
134 if (!update_fp_enabled(vcpu)) {
136 __activate_traps_fpsimd32(vcpu);
139 write_sysreg(val, cptr_el2);
141 if (cpus_have_final_cap(ARM64_WORKAROUND_SPECULATIVE_AT_NVHE)) {
142 struct kvm_cpu_context *ctxt = &vcpu->arch.ctxt;
146 * At this stage, and thanks to the above isb(), S2 is
147 * configured and enabled. We can now restore the guest's S1
148 * configuration: SCTLR, and only then TCR.
150 write_sysreg_el1(ctxt->sys_regs[SCTLR_EL1], SYS_SCTLR);
152 write_sysreg_el1(ctxt->sys_regs[TCR_EL1], SYS_TCR);
156 static void __hyp_text __activate_traps(struct kvm_vcpu *vcpu)
158 u64 hcr = vcpu->arch.hcr_el2;
160 if (cpus_have_final_cap(ARM64_WORKAROUND_CAVIUM_TX2_219_TVM))
163 write_sysreg(hcr, hcr_el2);
165 if (cpus_have_final_cap(ARM64_HAS_RAS_EXTN) && (hcr & HCR_VSE))
166 write_sysreg_s(vcpu->arch.vsesr_el2, SYS_VSESR_EL2);
169 activate_traps_vhe(vcpu);
171 __activate_traps_nvhe(vcpu);
174 static void deactivate_traps_vhe(void)
176 extern char vectors[]; /* kernel exception vectors */
177 write_sysreg(HCR_HOST_VHE_FLAGS, hcr_el2);
180 * ARM errata 1165522 and 1530923 require the actual execution of the
181 * above before we can switch to the EL2/EL0 translation regime used by
184 asm(ALTERNATIVE("nop", "isb", ARM64_WORKAROUND_SPECULATIVE_AT_VHE));
186 write_sysreg(CPACR_EL1_DEFAULT, cpacr_el1);
187 write_sysreg(vectors, vbar_el1);
189 NOKPROBE_SYMBOL(deactivate_traps_vhe);
191 static void __hyp_text __deactivate_traps_nvhe(void)
193 u64 mdcr_el2 = read_sysreg(mdcr_el2);
195 if (cpus_have_final_cap(ARM64_WORKAROUND_SPECULATIVE_AT_NVHE)) {
199 * Set the TCR and SCTLR registers in the exact opposite
200 * sequence as __activate_traps_nvhe (first prevent walks,
201 * then force the MMU on). A generous sprinkling of isb()
202 * ensure that things happen in this exact order.
204 val = read_sysreg_el1(SYS_TCR);
205 write_sysreg_el1(val | TCR_EPD1_MASK | TCR_EPD0_MASK, SYS_TCR);
207 val = read_sysreg_el1(SYS_SCTLR);
208 write_sysreg_el1(val | SCTLR_ELx_M, SYS_SCTLR);
212 __deactivate_traps_common();
214 mdcr_el2 &= MDCR_EL2_HPMN_MASK;
215 mdcr_el2 |= MDCR_EL2_E2PB_MASK << MDCR_EL2_E2PB_SHIFT;
217 write_sysreg(mdcr_el2, mdcr_el2);
218 write_sysreg(HCR_HOST_NVHE_FLAGS, hcr_el2);
219 write_sysreg(CPTR_EL2_DEFAULT, cptr_el2);
222 static void __hyp_text __deactivate_traps(struct kvm_vcpu *vcpu)
225 * If we pended a virtual abort, preserve it until it gets
226 * cleared. See D1.14.3 (Virtual Interrupts) for details, but
227 * the crucial bit is "On taking a vSError interrupt,
228 * HCR_EL2.VSE is cleared to 0."
230 if (vcpu->arch.hcr_el2 & HCR_VSE) {
231 vcpu->arch.hcr_el2 &= ~HCR_VSE;
232 vcpu->arch.hcr_el2 |= read_sysreg(hcr_el2) & HCR_VSE;
236 deactivate_traps_vhe();
238 __deactivate_traps_nvhe();
241 void activate_traps_vhe_load(struct kvm_vcpu *vcpu)
243 __activate_traps_common(vcpu);
246 void deactivate_traps_vhe_put(void)
248 u64 mdcr_el2 = read_sysreg(mdcr_el2);
250 mdcr_el2 &= MDCR_EL2_HPMN_MASK |
251 MDCR_EL2_E2PB_MASK << MDCR_EL2_E2PB_SHIFT |
254 write_sysreg(mdcr_el2, mdcr_el2);
256 __deactivate_traps_common();
259 static void __hyp_text __activate_vm(struct kvm *kvm)
261 __load_guest_stage2(kvm);
264 static void __hyp_text __deactivate_vm(struct kvm_vcpu *vcpu)
266 write_sysreg(0, vttbr_el2);
269 /* Save VGICv3 state on non-VHE systems */
270 static void __hyp_text __hyp_vgic_save_state(struct kvm_vcpu *vcpu)
272 if (static_branch_unlikely(&kvm_vgic_global_state.gicv3_cpuif)) {
273 __vgic_v3_save_state(vcpu);
274 __vgic_v3_deactivate_traps(vcpu);
278 /* Restore VGICv3 state on non_VEH systems */
279 static void __hyp_text __hyp_vgic_restore_state(struct kvm_vcpu *vcpu)
281 if (static_branch_unlikely(&kvm_vgic_global_state.gicv3_cpuif)) {
282 __vgic_v3_activate_traps(vcpu);
283 __vgic_v3_restore_state(vcpu);
287 static bool __hyp_text __translate_far_to_hpfar(u64 far, u64 *hpfar)
292 * Resolve the IPA the hard way using the guest VA.
294 * Stage-1 translation already validated the memory access
295 * rights. As such, we can use the EL1 translation regime, and
296 * don't have to distinguish between EL0 and EL1 access.
298 * We do need to save/restore PAR_EL1 though, as we haven't
299 * saved the guest context yet, and we may return early...
301 par = read_sysreg(par_el1);
302 asm volatile("at s1e1r, %0" : : "r" (far));
305 tmp = read_sysreg(par_el1);
306 write_sysreg(par, par_el1);
308 if (unlikely(tmp & SYS_PAR_EL1_F))
309 return false; /* Translation failed, back to guest */
311 /* Convert PAR to HPFAR format */
312 *hpfar = PAR_TO_HPFAR(tmp);
316 static bool __hyp_text __populate_fault_info(struct kvm_vcpu *vcpu)
322 esr = vcpu->arch.fault.esr_el2;
323 ec = ESR_ELx_EC(esr);
325 if (ec != ESR_ELx_EC_DABT_LOW && ec != ESR_ELx_EC_IABT_LOW)
328 far = read_sysreg_el2(SYS_FAR);
331 * The HPFAR can be invalid if the stage 2 fault did not
332 * happen during a stage 1 page table walk (the ESR_EL2.S1PTW
333 * bit is clear) and one of the two following cases are true:
334 * 1. The fault was due to a permission fault
335 * 2. The processor carries errata 834220
337 * Therefore, for all non S1PTW faults where we either have a
338 * permission fault or the errata workaround is enabled, we
339 * resolve the IPA using the AT instruction.
341 if (!(esr & ESR_ELx_S1PTW) &&
342 (cpus_have_final_cap(ARM64_WORKAROUND_834220) ||
343 (esr & ESR_ELx_FSC_TYPE) == FSC_PERM)) {
344 if (!__translate_far_to_hpfar(far, &hpfar))
347 hpfar = read_sysreg(hpfar_el2);
350 vcpu->arch.fault.far_el2 = far;
351 vcpu->arch.fault.hpfar_el2 = hpfar;
355 /* Check for an FPSIMD/SVE trap and handle as appropriate */
356 static bool __hyp_text __hyp_handle_fpsimd(struct kvm_vcpu *vcpu)
358 bool vhe, sve_guest, sve_host;
361 if (!system_supports_fpsimd())
364 if (system_supports_sve()) {
365 sve_guest = vcpu_has_sve(vcpu);
366 sve_host = vcpu->arch.flags & KVM_ARM64_HOST_SVE_IN_USE;
374 hsr_ec = kvm_vcpu_trap_get_class(vcpu);
375 if (hsr_ec != ESR_ELx_EC_FP_ASIMD &&
376 hsr_ec != ESR_ELx_EC_SVE)
379 /* Don't handle SVE traps for non-SVE vcpus here: */
381 if (hsr_ec != ESR_ELx_EC_FP_ASIMD)
384 /* Valid trap. Switch the context: */
387 u64 reg = read_sysreg(cpacr_el1) | CPACR_EL1_FPEN;
390 reg |= CPACR_EL1_ZEN;
392 write_sysreg(reg, cpacr_el1);
394 write_sysreg(read_sysreg(cptr_el2) & ~(u64)CPTR_EL2_TFP,
400 if (vcpu->arch.flags & KVM_ARM64_FP_HOST) {
402 * In the SVE case, VHE is assumed: it is enforced by
403 * Kconfig and kvm_arch_init().
406 struct thread_struct *thread = container_of(
407 vcpu->arch.host_fpsimd_state,
408 struct thread_struct, uw.fpsimd_state);
410 sve_save_state(sve_pffr(thread),
411 &vcpu->arch.host_fpsimd_state->fpsr);
413 __fpsimd_save_state(vcpu->arch.host_fpsimd_state);
416 vcpu->arch.flags &= ~KVM_ARM64_FP_HOST;
420 sve_load_state(vcpu_sve_pffr(vcpu),
421 &vcpu->arch.ctxt.gp_regs.fp_regs.fpsr,
422 sve_vq_from_vl(vcpu->arch.sve_max_vl) - 1);
423 write_sysreg_s(vcpu->arch.ctxt.sys_regs[ZCR_EL1], SYS_ZCR_EL12);
425 __fpsimd_restore_state(&vcpu->arch.ctxt.gp_regs.fp_regs);
428 /* Skip restoring fpexc32 for AArch64 guests */
429 if (!(read_sysreg(hcr_el2) & HCR_RW))
430 write_sysreg(vcpu->arch.ctxt.sys_regs[FPEXC32_EL2],
433 vcpu->arch.flags |= KVM_ARM64_FP_ENABLED;
438 static bool __hyp_text handle_tx2_tvm(struct kvm_vcpu *vcpu)
440 u32 sysreg = esr_sys64_to_sysreg(kvm_vcpu_get_hsr(vcpu));
441 int rt = kvm_vcpu_sys_get_rt(vcpu);
442 u64 val = vcpu_get_reg(vcpu, rt);
445 * The normal sysreg handling code expects to see the traps,
446 * let's not do anything here.
448 if (vcpu->arch.hcr_el2 & HCR_TVM)
453 write_sysreg_el1(val, SYS_SCTLR);
456 write_sysreg_el1(val, SYS_TTBR0);
459 write_sysreg_el1(val, SYS_TTBR1);
462 write_sysreg_el1(val, SYS_TCR);
465 write_sysreg_el1(val, SYS_ESR);
468 write_sysreg_el1(val, SYS_FAR);
471 write_sysreg_el1(val, SYS_AFSR0);
474 write_sysreg_el1(val, SYS_AFSR1);
477 write_sysreg_el1(val, SYS_MAIR);
480 write_sysreg_el1(val, SYS_AMAIR);
482 case SYS_CONTEXTIDR_EL1:
483 write_sysreg_el1(val, SYS_CONTEXTIDR);
489 __kvm_skip_instr(vcpu);
494 * Return true when we were able to fixup the guest exit and should return to
495 * the guest, false when we should restore the host state and return to the
498 static bool __hyp_text fixup_guest_exit(struct kvm_vcpu *vcpu, u64 *exit_code)
500 if (ARM_EXCEPTION_CODE(*exit_code) != ARM_EXCEPTION_IRQ)
501 vcpu->arch.fault.esr_el2 = read_sysreg_el2(SYS_ESR);
504 * We're using the raw exception code in order to only process
505 * the trap if no SError is pending. We will come back to the
506 * same PC once the SError has been injected, and replay the
507 * trapping instruction.
509 if (*exit_code != ARM_EXCEPTION_TRAP)
512 if (cpus_have_final_cap(ARM64_WORKAROUND_CAVIUM_TX2_219_TVM) &&
513 kvm_vcpu_trap_get_class(vcpu) == ESR_ELx_EC_SYS64 &&
514 handle_tx2_tvm(vcpu))
518 * We trap the first access to the FP/SIMD to save the host context
519 * and restore the guest context lazily.
520 * If FP/SIMD is not implemented, handle the trap and inject an
521 * undefined instruction exception to the guest.
522 * Similarly for trapped SVE accesses.
524 if (__hyp_handle_fpsimd(vcpu))
527 if (!__populate_fault_info(vcpu))
530 if (static_branch_unlikely(&vgic_v2_cpuif_trap)) {
533 valid = kvm_vcpu_trap_get_class(vcpu) == ESR_ELx_EC_DABT_LOW &&
534 kvm_vcpu_trap_get_fault_type(vcpu) == FSC_FAULT &&
535 kvm_vcpu_dabt_isvalid(vcpu) &&
536 !kvm_vcpu_dabt_isextabt(vcpu) &&
537 !kvm_vcpu_dabt_iss1tw(vcpu);
540 int ret = __vgic_v2_perform_cpuif_access(vcpu);
545 /* Promote an illegal access to an SError.*/
547 *exit_code = ARM_EXCEPTION_EL1_SERROR;
553 if (static_branch_unlikely(&vgic_v3_cpuif_trap) &&
554 (kvm_vcpu_trap_get_class(vcpu) == ESR_ELx_EC_SYS64 ||
555 kvm_vcpu_trap_get_class(vcpu) == ESR_ELx_EC_CP15_32)) {
556 int ret = __vgic_v3_perform_cpuif_access(vcpu);
563 /* Return to the host kernel and handle the exit */
567 static inline bool __hyp_text __needs_ssbd_off(struct kvm_vcpu *vcpu)
569 if (!cpus_have_final_cap(ARM64_SSBD))
572 return !(vcpu->arch.workaround_flags & VCPU_WORKAROUND_2_FLAG);
575 static void __hyp_text __set_guest_arch_workaround_state(struct kvm_vcpu *vcpu)
577 #ifdef CONFIG_ARM64_SSBD
579 * The host runs with the workaround always present. If the
580 * guest wants it disabled, so be it...
582 if (__needs_ssbd_off(vcpu) &&
583 __hyp_this_cpu_read(arm64_ssbd_callback_required))
584 arm_smccc_1_1_smc(ARM_SMCCC_ARCH_WORKAROUND_2, 0, NULL);
588 static void __hyp_text __set_host_arch_workaround_state(struct kvm_vcpu *vcpu)
590 #ifdef CONFIG_ARM64_SSBD
592 * If the guest has disabled the workaround, bring it back on.
594 if (__needs_ssbd_off(vcpu) &&
595 __hyp_this_cpu_read(arm64_ssbd_callback_required))
596 arm_smccc_1_1_smc(ARM_SMCCC_ARCH_WORKAROUND_2, 1, NULL);
601 * Disable host events, enable guest events
603 static bool __hyp_text __pmu_switch_to_guest(struct kvm_cpu_context *host_ctxt)
605 struct kvm_host_data *host;
606 struct kvm_pmu_events *pmu;
608 host = container_of(host_ctxt, struct kvm_host_data, host_ctxt);
609 pmu = &host->pmu_events;
611 if (pmu->events_host)
612 write_sysreg(pmu->events_host, pmcntenclr_el0);
614 if (pmu->events_guest)
615 write_sysreg(pmu->events_guest, pmcntenset_el0);
617 return (pmu->events_host || pmu->events_guest);
621 * Disable guest events, enable host events
623 static void __hyp_text __pmu_switch_to_host(struct kvm_cpu_context *host_ctxt)
625 struct kvm_host_data *host;
626 struct kvm_pmu_events *pmu;
628 host = container_of(host_ctxt, struct kvm_host_data, host_ctxt);
629 pmu = &host->pmu_events;
631 if (pmu->events_guest)
632 write_sysreg(pmu->events_guest, pmcntenclr_el0);
634 if (pmu->events_host)
635 write_sysreg(pmu->events_host, pmcntenset_el0);
638 /* Switch to the guest for VHE systems running in EL2 */
639 static int __kvm_vcpu_run_vhe(struct kvm_vcpu *vcpu)
641 struct kvm_cpu_context *host_ctxt;
642 struct kvm_cpu_context *guest_ctxt;
645 host_ctxt = vcpu->arch.host_cpu_context;
646 host_ctxt->__hyp_running_vcpu = vcpu;
647 guest_ctxt = &vcpu->arch.ctxt;
649 sysreg_save_host_state_vhe(host_ctxt);
652 * ARM erratum 1165522 requires us to configure both stage 1 and
653 * stage 2 translation for the guest context before we clear
656 * We have already configured the guest's stage 1 translation in
657 * kvm_vcpu_load_sysregs above. We must now call __activate_vm
658 * before __activate_traps, because __activate_vm configures
659 * stage 2 translation, and __activate_traps clear HCR_EL2.TGE
660 * (among other things).
662 __activate_vm(vcpu->kvm);
663 __activate_traps(vcpu);
665 sysreg_restore_guest_state_vhe(guest_ctxt);
666 __debug_switch_to_guest(vcpu);
668 __set_guest_arch_workaround_state(vcpu);
671 /* Jump in the fire! */
672 exit_code = __guest_enter(vcpu, host_ctxt);
674 /* And we're baaack! */
675 } while (fixup_guest_exit(vcpu, &exit_code));
677 __set_host_arch_workaround_state(vcpu);
679 sysreg_save_guest_state_vhe(guest_ctxt);
681 __deactivate_traps(vcpu);
683 sysreg_restore_host_state_vhe(host_ctxt);
685 if (vcpu->arch.flags & KVM_ARM64_FP_ENABLED)
686 __fpsimd_save_fpexc32(vcpu);
688 __debug_switch_to_host(vcpu);
692 NOKPROBE_SYMBOL(__kvm_vcpu_run_vhe);
694 int kvm_vcpu_run_vhe(struct kvm_vcpu *vcpu)
701 * Having IRQs masked via PMR when entering the guest means the GIC
702 * will not signal the CPU of interrupts of lower priority, and the
703 * only way to get out will be via guest exceptions.
704 * Naturally, we want to avoid this.
706 * local_daif_mask() already sets GIC_PRIO_PSR_I_SET, we just need a
707 * dsb to ensure the redistributor is forwards EL2 IRQs to the CPU.
711 ret = __kvm_vcpu_run_vhe(vcpu);
714 * local_daif_restore() takes care to properly restore PSTATE.DAIF
715 * and the GIC PMR if the host is using IRQ priorities.
717 local_daif_restore(DAIF_PROCCTX_NOIRQ);
720 * When we exit from the guest we change a number of CPU configuration
721 * parameters, such as traps. Make sure these changes take effect
722 * before running the host or additional guests.
729 /* Switch to the guest for legacy non-VHE systems */
730 int __hyp_text __kvm_vcpu_run_nvhe(struct kvm_vcpu *vcpu)
732 struct kvm_cpu_context *host_ctxt;
733 struct kvm_cpu_context *guest_ctxt;
734 bool pmu_switch_needed;
738 * Having IRQs masked via PMR when entering the guest means the GIC
739 * will not signal the CPU of interrupts of lower priority, and the
740 * only way to get out will be via guest exceptions.
741 * Naturally, we want to avoid this.
743 if (system_uses_irq_prio_masking()) {
744 gic_write_pmr(GIC_PRIO_IRQON | GIC_PRIO_PSR_I_SET);
748 vcpu = kern_hyp_va(vcpu);
750 host_ctxt = kern_hyp_va(vcpu->arch.host_cpu_context);
751 host_ctxt->__hyp_running_vcpu = vcpu;
752 guest_ctxt = &vcpu->arch.ctxt;
754 pmu_switch_needed = __pmu_switch_to_guest(host_ctxt);
756 __sysreg_save_state_nvhe(host_ctxt);
759 * We must restore the 32-bit state before the sysregs, thanks
760 * to erratum #852523 (Cortex-A57) or #853709 (Cortex-A72).
762 * Also, and in order to be able to deal with erratum #1319537 (A57)
763 * and #1319367 (A72), we must ensure that all VM-related sysreg are
764 * restored before we enable S2 translation.
766 __sysreg32_restore_state(vcpu);
767 __sysreg_restore_state_nvhe(guest_ctxt);
769 __activate_vm(kern_hyp_va(vcpu->kvm));
770 __activate_traps(vcpu);
772 __hyp_vgic_restore_state(vcpu);
773 __timer_enable_traps(vcpu);
775 __debug_switch_to_guest(vcpu);
777 __set_guest_arch_workaround_state(vcpu);
780 /* Jump in the fire! */
781 exit_code = __guest_enter(vcpu, host_ctxt);
783 /* And we're baaack! */
784 } while (fixup_guest_exit(vcpu, &exit_code));
786 __set_host_arch_workaround_state(vcpu);
788 __sysreg_save_state_nvhe(guest_ctxt);
789 __sysreg32_save_state(vcpu);
790 __timer_disable_traps(vcpu);
791 __hyp_vgic_save_state(vcpu);
793 __deactivate_traps(vcpu);
794 __deactivate_vm(vcpu);
796 __sysreg_restore_state_nvhe(host_ctxt);
798 if (vcpu->arch.flags & KVM_ARM64_FP_ENABLED)
799 __fpsimd_save_fpexc32(vcpu);
802 * This must come after restoring the host sysregs, since a non-VHE
803 * system may enable SPE here and make use of the TTBRs.
805 __debug_switch_to_host(vcpu);
807 if (pmu_switch_needed)
808 __pmu_switch_to_host(host_ctxt);
810 /* Returning to host will clear PSR.I, remask PMR if needed */
811 if (system_uses_irq_prio_masking())
812 gic_write_pmr(GIC_PRIO_IRQOFF);
817 static const char __hyp_panic_string[] = "HYP panic:\nPS:%08llx PC:%016llx ESR:%08llx\nFAR:%016llx HPFAR:%016llx PAR:%016llx\nVCPU:%p\n";
819 static void __hyp_text __hyp_call_panic_nvhe(u64 spsr, u64 elr, u64 par,
820 struct kvm_cpu_context *__host_ctxt)
822 struct kvm_vcpu *vcpu;
823 unsigned long str_va;
825 vcpu = __host_ctxt->__hyp_running_vcpu;
827 if (read_sysreg(vttbr_el2)) {
828 __timer_disable_traps(vcpu);
829 __deactivate_traps(vcpu);
830 __deactivate_vm(vcpu);
831 __sysreg_restore_state_nvhe(__host_ctxt);
835 * Force the panic string to be loaded from the literal pool,
836 * making sure it is a kernel address and not a PC-relative
839 asm volatile("ldr %0, =__hyp_panic_string" : "=r" (str_va));
841 __hyp_do_panic(str_va,
843 read_sysreg(esr_el2), read_sysreg_el2(SYS_FAR),
844 read_sysreg(hpfar_el2), par, vcpu);
847 static void __hyp_call_panic_vhe(u64 spsr, u64 elr, u64 par,
848 struct kvm_cpu_context *host_ctxt)
850 struct kvm_vcpu *vcpu;
851 vcpu = host_ctxt->__hyp_running_vcpu;
853 __deactivate_traps(vcpu);
854 sysreg_restore_host_state_vhe(host_ctxt);
856 panic(__hyp_panic_string,
858 read_sysreg_el2(SYS_ESR), read_sysreg_el2(SYS_FAR),
859 read_sysreg(hpfar_el2), par, vcpu);
861 NOKPROBE_SYMBOL(__hyp_call_panic_vhe);
863 void __hyp_text __noreturn hyp_panic(struct kvm_cpu_context *host_ctxt)
865 u64 spsr = read_sysreg_el2(SYS_SPSR);
866 u64 elr = read_sysreg_el2(SYS_ELR);
867 u64 par = read_sysreg(par_el1);
870 __hyp_call_panic_nvhe(spsr, elr, par, host_ctxt);
872 __hyp_call_panic_vhe(spsr, elr, par, host_ctxt);